git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/fs-io-pagecache.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #ifndef NO_BCACHEFS_FS
   3
   4 #include "bcachefs.h"
   5 #include "btree_iter.h"
   6 #include "extents.h"
   7 #include "fs-io.h"
   8 #include "fs-io-pagecache.h"
   9 #include "subvolume.h"
  10
  11 #include <linux/pagevec.h>
  12 #include <linux/writeback.h>
  13
  14 int bch2_filemap_get_contig_folios_d(struct address_space *mapping,
  15                                      loff_t start, u64 end,
  16                                      int fgp_flags, gfp_t gfp,
  17                                      folios *folios)
  18 {
  19         struct folio *f;
  20         u64 pos = start;
  21         int ret = 0;
  22
  23         while (pos < end) {
  24                 if ((u64) pos >= (u64) start + (1ULL << 20))
  25                         fgp_flags &= ~FGP_CREAT;
  26
  27                 ret = darray_make_room_gfp(folios, 1, gfp & GFP_KERNEL);
  28                 if (ret)
  29                         break;
  30
  31                 f = __filemap_get_folio(mapping, pos >> PAGE_SHIFT, fgp_flags, gfp);
  32                 if (IS_ERR_OR_NULL(f))
  33                         break;
  34
  35                 BUG_ON(folios->nr && folio_pos(f) != pos);
  36
  37                 pos = folio_end_pos(f);
  38                 darray_push(folios, f);
  39         }
  40
  41         if (!folios->nr && !ret && (fgp_flags & FGP_CREAT))
  42                 ret = -ENOMEM;
  43
  44         return folios->nr ? 0 : ret;
  45 }
  46
  47 /* pagecache_block must be held */
  48 int bch2_write_invalidate_inode_pages_range(struct address_space *mapping,
  49                                             loff_t start, loff_t end)
  50 {
  51         int ret;
  52
  53         /*
  54          * XXX: the way this is currently implemented, we can spin if a process
  55          * is continually redirtying a specific page
  56          */
  57         do {
  58                 if (!mapping->nrpages)
  59                         return 0;
  60
  61                 ret = filemap_write_and_wait_range(mapping, start, end);
  62                 if (ret)
  63                         break;
  64
  65                 if (!mapping->nrpages)
  66                         return 0;
  67
  68                 ret = invalidate_inode_pages2_range(mapping,
  69                                 start >> PAGE_SHIFT,
  70                                 end >> PAGE_SHIFT);
  71         } while (ret == -EBUSY);
  72
  73         return ret;
  74 }
  75
  76 static const char * const bch2_folio_sector_states[] = {
  77 #define x(n)    #n,
  78         BCH_FOLIO_SECTOR_STATE()
  79 #undef x
  80         NULL
  81 };
  82
  83 static inline enum bch_folio_sector_state
  84 folio_sector_dirty(enum bch_folio_sector_state state)
  85 {
  86         switch (state) {
  87         case SECTOR_unallocated:
  88                 return SECTOR_dirty;
  89         case SECTOR_reserved:
  90                 return SECTOR_dirty_reserved;
  91         default:
  92                 return state;
  93         }
  94 }
  95
  96 static inline enum bch_folio_sector_state
  97 folio_sector_undirty(enum bch_folio_sector_state state)
  98 {
  99         switch (state) {
 100         case SECTOR_dirty:
 101                 return SECTOR_unallocated;
 102         case SECTOR_dirty_reserved:
 103                 return SECTOR_reserved;
 104         default:
 105                 return state;
 106         }
 107 }
 108
 109 static inline enum bch_folio_sector_state
 110 folio_sector_reserve(enum bch_folio_sector_state state)
 111 {
 112         switch (state) {
 113         case SECTOR_unallocated:
 114                 return SECTOR_reserved;
 115         case SECTOR_dirty:
 116                 return SECTOR_dirty_reserved;
 117         default:
 118                 return state;
 119         }
 120 }
 121
 122 /* for newly allocated folios: */
 123 struct bch_folio *__bch2_folio_create(struct folio *folio, gfp_t gfp)
 124 {
 125         struct bch_folio *s;
 126
 127         s = kzalloc(sizeof(*s) +
 128                     sizeof(struct bch_folio_sector) *
 129                     folio_sectors(folio), gfp);
 130         if (!s)
 131                 return NULL;
 132
 133         spin_lock_init(&s->lock);
 134         folio_attach_private(folio, s);
 135         return s;
 136 }
 137
 138 struct bch_folio *bch2_folio_create(struct folio *folio, gfp_t gfp)
 139 {
 140         return bch2_folio(folio) ?: __bch2_folio_create(folio, gfp);
 141 }
 142
 143 static unsigned bkey_to_sector_state(struct bkey_s_c k)
 144 {
 145         if (bkey_extent_is_reservation(k))
 146                 return SECTOR_reserved;
 147         if (bkey_extent_is_allocation(k.k))
 148                 return SECTOR_allocated;
 149         return SECTOR_unallocated;
 150 }
 151
 152 static void __bch2_folio_set(struct folio *folio,
 153                              unsigned pg_offset, unsigned pg_len,
 154                              unsigned nr_ptrs, unsigned state)
 155 {
 156         struct bch_folio *s = bch2_folio(folio);
 157         unsigned i, sectors = folio_sectors(folio);
 158
 159         BUG_ON(pg_offset >= sectors);
 160         BUG_ON(pg_offset + pg_len > sectors);
 161
 162         spin_lock(&s->lock);
 163
 164         for (i = pg_offset; i < pg_offset + pg_len; i++) {
 165                 s->s[i].nr_replicas     = nr_ptrs;
 166                 bch2_folio_sector_set(folio, s, i, state);
 167         }
 168
 169         if (i == sectors)
 170                 s->uptodate = true;
 171
 172         spin_unlock(&s->lock);
 173 }
 174
 175 /*
 176  * Initialize bch_folio state (allocated/unallocated, nr_replicas) from the
 177  * extents btree:
 178  */
 179 int bch2_folio_set(struct bch_fs *c, subvol_inum inum,
 180                    struct folio **folios, unsigned nr_folios)
 181 {
 182         struct btree_trans trans;
 183         struct btree_iter iter;
 184         struct bkey_s_c k;
 185         struct bch_folio *s;
 186         u64 offset = folio_sector(folios[0]);
 187         unsigned folio_idx;
 188         u32 snapshot;
 189         bool need_set = false;
 190         int ret;
 191
 192         for (folio_idx = 0; folio_idx < nr_folios; folio_idx++) {
 193                 s = bch2_folio_create(folios[folio_idx], GFP_KERNEL);
 194                 if (!s)
 195                         return -ENOMEM;
 196
 197                 need_set |= !s->uptodate;
 198         }
 199
 200         if (!need_set)
 201                 return 0;
 202
 203         folio_idx = 0;
 204         bch2_trans_init(&trans, c, 0, 0);
 205 retry:
 206         bch2_trans_begin(&trans);
 207
 208         ret = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot);
 209         if (ret)
 210                 goto err;
 211
 212         for_each_btree_key_norestart(&trans, iter, BTREE_ID_extents,
 213                            SPOS(inum.inum, offset, snapshot),
 214                            BTREE_ITER_SLOTS, k, ret) {
 215                 unsigned nr_ptrs = bch2_bkey_nr_ptrs_fully_allocated(k);
 216                 unsigned state = bkey_to_sector_state(k);
 217
 218                 while (folio_idx < nr_folios) {
 219                         struct folio *folio = folios[folio_idx];
 220                         u64 folio_start = folio_sector(folio);
 221                         u64 folio_end   = folio_end_sector(folio);
 222                         unsigned folio_offset = max(bkey_start_offset(k.k), folio_start) - folio_start;
 223                         unsigned folio_len = min(k.k->p.offset, folio_end) - folio_offset - folio_start;
 224
 225                         BUG_ON(k.k->p.offset < folio_start);
 226                         BUG_ON(bkey_start_offset(k.k) > folio_end);
 227
 228                         if (!bch2_folio(folio)->uptodate)
 229                                 __bch2_folio_set(folio, folio_offset, folio_len, nr_ptrs, state);
 230
 231                         if (k.k->p.offset < folio_end)
 232                                 break;
 233                         folio_idx++;
 234                 }
 235
 236                 if (folio_idx == nr_folios)
 237                         break;
 238         }
 239
 240         offset = iter.pos.offset;
 241         bch2_trans_iter_exit(&trans, &iter);
 242 err:
 243         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 244                 goto retry;
 245         bch2_trans_exit(&trans);
 246
 247         return ret;
 248 }
 249
 250 void bch2_bio_page_state_set(struct bio *bio, struct bkey_s_c k)
 251 {
 252         struct bvec_iter iter;
 253         struct folio_vec fv;
 254         unsigned nr_ptrs = k.k->type == KEY_TYPE_reflink_v
 255                 ? 0 : bch2_bkey_nr_ptrs_fully_allocated(k);
 256         unsigned state = bkey_to_sector_state(k);
 257
 258         bio_for_each_folio(fv, bio, iter)
 259                 __bch2_folio_set(fv.fv_folio,
 260                                  fv.fv_offset >> 9,
 261                                  fv.fv_len >> 9,
 262                                  nr_ptrs, state);
 263 }
 264
 265 void bch2_mark_pagecache_unallocated(struct bch_inode_info *inode,
 266                                      u64 start, u64 end)
 267 {
 268         pgoff_t index = start >> PAGE_SECTORS_SHIFT;
 269         pgoff_t end_index = (end - 1) >> PAGE_SECTORS_SHIFT;
 270         struct folio_batch fbatch;
 271         unsigned i, j;
 272
 273         if (end <= start)
 274                 return;
 275
 276         folio_batch_init(&fbatch);
 277
 278         while (filemap_get_folios(inode->v.i_mapping,
 279                                   &index, end_index, &fbatch)) {
 280                 for (i = 0; i < folio_batch_count(&fbatch); i++) {
 281                         struct folio *folio = fbatch.folios[i];
 282                         u64 folio_start = folio_sector(folio);
 283                         u64 folio_end = folio_end_sector(folio);
 284                         unsigned folio_offset = max(start, folio_start) - folio_start;
 285                         unsigned folio_len = min(end, folio_end) - folio_offset - folio_start;
 286                         struct bch_folio *s;
 287
 288                         BUG_ON(end <= folio_start);
 289
 290                         folio_lock(folio);
 291                         s = bch2_folio(folio);
 292
 293                         if (s) {
 294                                 spin_lock(&s->lock);
 295                                 for (j = folio_offset; j < folio_offset + folio_len; j++)
 296                                         s->s[j].nr_replicas = 0;
 297                                 spin_unlock(&s->lock);
 298                         }
 299
 300                         folio_unlock(folio);
 301                 }
 302                 folio_batch_release(&fbatch);
 303                 cond_resched();
 304         }
 305 }
 306
 307 void bch2_mark_pagecache_reserved(struct bch_inode_info *inode,
 308                                   u64 start, u64 end)
 309 {
 310         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 311         pgoff_t index = start >> PAGE_SECTORS_SHIFT;
 312         pgoff_t end_index = (end - 1) >> PAGE_SECTORS_SHIFT;
 313         struct folio_batch fbatch;
 314         s64 i_sectors_delta = 0;
 315         unsigned i, j;
 316
 317         if (end <= start)
 318                 return;
 319
 320         folio_batch_init(&fbatch);
 321
 322         while (filemap_get_folios(inode->v.i_mapping,
 323                                   &index, end_index, &fbatch)) {
 324                 for (i = 0; i < folio_batch_count(&fbatch); i++) {
 325                         struct folio *folio = fbatch.folios[i];
 326                         u64 folio_start = folio_sector(folio);
 327                         u64 folio_end = folio_end_sector(folio);
 328                         unsigned folio_offset = max(start, folio_start) - folio_start;
 329                         unsigned folio_len = min(end, folio_end) - folio_offset - folio_start;
 330                         struct bch_folio *s;
 331
 332                         BUG_ON(end <= folio_start);
 333
 334                         folio_lock(folio);
 335                         s = bch2_folio(folio);
 336
 337                         if (s) {
 338                                 spin_lock(&s->lock);
 339                                 for (j = folio_offset; j < folio_offset + folio_len; j++) {
 340                                         i_sectors_delta -= s->s[j].state == SECTOR_dirty;
 341                                         bch2_folio_sector_set(folio, s, j, folio_sector_reserve(s->s[j].state));
 342                                 }
 343                                 spin_unlock(&s->lock);
 344                         }
 345
 346                         folio_unlock(folio);
 347                 }
 348                 folio_batch_release(&fbatch);
 349                 cond_resched();
 350         }
 351
 352         bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
 353 }
 354
 355 static inline unsigned sectors_to_reserve(struct bch_folio_sector *s,
 356                                           unsigned nr_replicas)
 357 {
 358         return max(0, (int) nr_replicas -
 359                    s->nr_replicas -
 360                    s->replicas_reserved);
 361 }
 362
 363 int bch2_get_folio_disk_reservation(struct bch_fs *c,
 364                                 struct bch_inode_info *inode,
 365                                 struct folio *folio, bool check_enospc)
 366 {
 367         struct bch_folio *s = bch2_folio_create(folio, 0);
 368         unsigned nr_replicas = inode_nr_replicas(c, inode);
 369         struct disk_reservation disk_res = { 0 };
 370         unsigned i, sectors = folio_sectors(folio), disk_res_sectors = 0;
 371         int ret;
 372
 373         if (!s)
 374                 return -ENOMEM;
 375
 376         for (i = 0; i < sectors; i++)
 377                 disk_res_sectors += sectors_to_reserve(&s->s[i], nr_replicas);
 378
 379         if (!disk_res_sectors)
 380                 return 0;
 381
 382         ret = bch2_disk_reservation_get(c, &disk_res,
 383                                         disk_res_sectors, 1,
 384                                         !check_enospc
 385                                         ? BCH_DISK_RESERVATION_NOFAIL
 386                                         : 0);
 387         if (unlikely(ret))
 388                 return ret;
 389
 390         for (i = 0; i < sectors; i++)
 391                 s->s[i].replicas_reserved +=
 392                         sectors_to_reserve(&s->s[i], nr_replicas);
 393
 394         return 0;
 395 }
 396
 397 void bch2_folio_reservation_put(struct bch_fs *c,
 398                         struct bch_inode_info *inode,
 399                         struct bch2_folio_reservation *res)
 400 {
 401         bch2_disk_reservation_put(c, &res->disk);
 402         bch2_quota_reservation_put(c, inode, &res->quota);
 403 }
 404
 405 int bch2_folio_reservation_get(struct bch_fs *c,
 406                         struct bch_inode_info *inode,
 407                         struct folio *folio,
 408                         struct bch2_folio_reservation *res,
 409                         unsigned offset, unsigned len)
 410 {
 411         struct bch_folio *s = bch2_folio_create(folio, 0);
 412         unsigned i, disk_sectors = 0, quota_sectors = 0;
 413         int ret;
 414
 415         if (!s)
 416                 return -ENOMEM;
 417
 418         BUG_ON(!s->uptodate);
 419
 420         for (i = round_down(offset, block_bytes(c)) >> 9;
 421              i < round_up(offset + len, block_bytes(c)) >> 9;
 422              i++) {
 423                 disk_sectors += sectors_to_reserve(&s->s[i],
 424                                                 res->disk.nr_replicas);
 425                 quota_sectors += s->s[i].state == SECTOR_unallocated;
 426         }
 427
 428         if (disk_sectors) {
 429                 ret = bch2_disk_reservation_add(c, &res->disk, disk_sectors, 0);
 430                 if (unlikely(ret))
 431                         return ret;
 432         }
 433
 434         if (quota_sectors) {
 435                 ret = bch2_quota_reservation_add(c, inode, &res->quota,
 436                                                  quota_sectors, true);
 437                 if (unlikely(ret)) {
 438                         struct disk_reservation tmp = {
 439                                 .sectors = disk_sectors
 440                         };
 441
 442                         bch2_disk_reservation_put(c, &tmp);
 443                         res->disk.sectors -= disk_sectors;
 444                         return ret;
 445                 }
 446         }
 447
 448         return 0;
 449 }
 450
 451 static void bch2_clear_folio_bits(struct folio *folio)
 452 {
 453         struct bch_inode_info *inode = to_bch_ei(folio->mapping->host);
 454         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 455         struct bch_folio *s = bch2_folio(folio);
 456         struct disk_reservation disk_res = { 0 };
 457         int i, sectors = folio_sectors(folio), dirty_sectors = 0;
 458
 459         if (!s)
 460                 return;
 461
 462         EBUG_ON(!folio_test_locked(folio));
 463         EBUG_ON(folio_test_writeback(folio));
 464
 465         for (i = 0; i < sectors; i++) {
 466                 disk_res.sectors += s->s[i].replicas_reserved;
 467                 s->s[i].replicas_reserved = 0;
 468
 469                 dirty_sectors -= s->s[i].state == SECTOR_dirty;
 470                 bch2_folio_sector_set(folio, s, i, folio_sector_undirty(s->s[i].state));
 471         }
 472
 473         bch2_disk_reservation_put(c, &disk_res);
 474
 475         bch2_i_sectors_acct(c, inode, NULL, dirty_sectors);
 476
 477         bch2_folio_release(folio);
 478 }
 479
 480 void bch2_set_folio_dirty(struct bch_fs *c,
 481                           struct bch_inode_info *inode,
 482                           struct folio *folio,
 483                           struct bch2_folio_reservation *res,
 484                           unsigned offset, unsigned len)
 485 {
 486         struct bch_folio *s = bch2_folio(folio);
 487         unsigned i, dirty_sectors = 0;
 488
 489         WARN_ON((u64) folio_pos(folio) + offset + len >
 490                 round_up((u64) i_size_read(&inode->v), block_bytes(c)));
 491
 492         BUG_ON(!s->uptodate);
 493
 494         spin_lock(&s->lock);
 495
 496         for (i = round_down(offset, block_bytes(c)) >> 9;
 497              i < round_up(offset + len, block_bytes(c)) >> 9;
 498              i++) {
 499                 unsigned sectors = sectors_to_reserve(&s->s[i],
 500                                                 res->disk.nr_replicas);
 501
 502                 /*
 503                  * This can happen if we race with the error path in
 504                  * bch2_writepage_io_done():
 505                  */
 506                 sectors = min_t(unsigned, sectors, res->disk.sectors);
 507
 508                 s->s[i].replicas_reserved += sectors;
 509                 res->disk.sectors -= sectors;
 510
 511                 dirty_sectors += s->s[i].state == SECTOR_unallocated;
 512
 513                 bch2_folio_sector_set(folio, s, i, folio_sector_dirty(s->s[i].state));
 514         }
 515
 516         spin_unlock(&s->lock);
 517
 518         bch2_i_sectors_acct(c, inode, &res->quota, dirty_sectors);
 519
 520         if (!folio_test_dirty(folio))
 521                 filemap_dirty_folio(inode->v.i_mapping, folio);
 522 }
 523
 524 vm_fault_t bch2_page_fault(struct vm_fault *vmf)
 525 {
 526         struct file *file = vmf->vma->vm_file;
 527         struct address_space *mapping = file->f_mapping;
 528         struct address_space *fdm = faults_disabled_mapping();
 529         struct bch_inode_info *inode = file_bch_inode(file);
 530         vm_fault_t ret;
 531
 532         if (fdm == mapping)
 533                 return VM_FAULT_SIGBUS;
 534
 535         /* Lock ordering: */
 536         if (fdm > mapping) {
 537                 struct bch_inode_info *fdm_host = to_bch_ei(fdm->host);
 538
 539                 if (bch2_pagecache_add_tryget(inode))
 540                         goto got_lock;
 541
 542                 bch2_pagecache_block_put(fdm_host);
 543
 544                 bch2_pagecache_add_get(inode);
 545                 bch2_pagecache_add_put(inode);
 546
 547                 bch2_pagecache_block_get(fdm_host);
 548
 549                 /* Signal that lock has been dropped: */
 550                 set_fdm_dropped_locks();
 551                 return VM_FAULT_SIGBUS;
 552         }
 553
 554         bch2_pagecache_add_get(inode);
 555 got_lock:
 556         ret = filemap_fault(vmf);
 557         bch2_pagecache_add_put(inode);
 558
 559         return ret;
 560 }
 561
 562 vm_fault_t bch2_page_mkwrite(struct vm_fault *vmf)
 563 {
 564         struct folio *folio = page_folio(vmf->page);
 565         struct file *file = vmf->vma->vm_file;
 566         struct bch_inode_info *inode = file_bch_inode(file);
 567         struct address_space *mapping = file->f_mapping;
 568         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 569         struct bch2_folio_reservation res;
 570         unsigned len;
 571         loff_t isize;
 572         vm_fault_t ret;
 573
 574         bch2_folio_reservation_init(c, inode, &res);
 575
 576         sb_start_pagefault(inode->v.i_sb);
 577         file_update_time(file);
 578
 579         /*
 580          * Not strictly necessary, but helps avoid dio writes livelocking in
 581          * bch2_write_invalidate_inode_pages_range() - can drop this if/when we get
 582          * a bch2_write_invalidate_inode_pages_range() that works without dropping
 583          * page lock before invalidating page
 584          */
 585         bch2_pagecache_add_get(inode);
 586
 587         folio_lock(folio);
 588         isize = i_size_read(&inode->v);
 589
 590         if (folio->mapping != mapping || folio_pos(folio) >= isize) {
 591                 folio_unlock(folio);
 592                 ret = VM_FAULT_NOPAGE;
 593                 goto out;
 594         }
 595
 596         len = min_t(loff_t, folio_size(folio), isize - folio_pos(folio));
 597
 598         if (bch2_folio_set(c, inode_inum(inode), &folio, 1) ?:
 599             bch2_folio_reservation_get(c, inode, folio, &res, 0, len)) {
 600                 folio_unlock(folio);
 601                 ret = VM_FAULT_SIGBUS;
 602                 goto out;
 603         }
 604
 605         bch2_set_folio_dirty(c, inode, folio, &res, 0, len);
 606         bch2_folio_reservation_put(c, inode, &res);
 607
 608         folio_wait_stable(folio);
 609         ret = VM_FAULT_LOCKED;
 610 out:
 611         bch2_pagecache_add_put(inode);
 612         sb_end_pagefault(inode->v.i_sb);
 613
 614         return ret;
 615 }
 616
 617 void bch2_invalidate_folio(struct folio *folio, size_t offset, size_t length)
 618 {
 619         if (offset || length < folio_size(folio))
 620                 return;
 621
 622         bch2_clear_folio_bits(folio);
 623 }
 624
 625 bool bch2_release_folio(struct folio *folio, gfp_t gfp_mask)
 626 {
 627         if (folio_test_dirty(folio) || folio_test_writeback(folio))
 628                 return false;
 629
 630         bch2_clear_folio_bits(folio);
 631         return true;
 632 }
 633
 634 /* fseek: */
 635
 636 static int folio_data_offset(struct folio *folio, loff_t pos,
 637                              unsigned min_replicas)
 638 {
 639         struct bch_folio *s = bch2_folio(folio);
 640         unsigned i, sectors = folio_sectors(folio);
 641
 642         if (s)
 643                 for (i = folio_pos_to_s(folio, pos); i < sectors; i++)
 644                         if (s->s[i].state >= SECTOR_dirty &&
 645                             s->s[i].nr_replicas + s->s[i].replicas_reserved >= min_replicas)
 646                                 return i << SECTOR_SHIFT;
 647
 648         return -1;
 649 }
 650
 651 loff_t bch2_seek_pagecache_data(struct inode *vinode,
 652                                 loff_t start_offset,
 653                                 loff_t end_offset,
 654                                 unsigned min_replicas,
 655                                 bool nonblock)
 656 {
 657         struct folio_batch fbatch;
 658         pgoff_t start_index     = start_offset >> PAGE_SHIFT;
 659         pgoff_t end_index       = end_offset >> PAGE_SHIFT;
 660         pgoff_t index           = start_index;
 661         unsigned i;
 662         loff_t ret;
 663         int offset;
 664
 665         folio_batch_init(&fbatch);
 666
 667         while (filemap_get_folios(vinode->i_mapping,
 668                                   &index, end_index, &fbatch)) {
 669                 for (i = 0; i < folio_batch_count(&fbatch); i++) {
 670                         struct folio *folio = fbatch.folios[i];
 671
 672                         if (!nonblock) {
 673                                 folio_lock(folio);
 674                         } else if (!folio_trylock(folio)) {
 675                                 folio_batch_release(&fbatch);
 676                                 return -EAGAIN;
 677                         }
 678
 679                         offset = folio_data_offset(folio,
 680                                         max(folio_pos(folio), start_offset),
 681                                         min_replicas);
 682                         if (offset >= 0) {
 683                                 ret = clamp(folio_pos(folio) + offset,
 684                                             start_offset, end_offset);
 685                                 folio_unlock(folio);
 686                                 folio_batch_release(&fbatch);
 687                                 return ret;
 688                         }
 689                         folio_unlock(folio);
 690                 }
 691                 folio_batch_release(&fbatch);
 692                 cond_resched();
 693         }
 694
 695         return end_offset;
 696 }
 697
 698 static int folio_hole_offset(struct address_space *mapping, loff_t *offset,
 699                               unsigned min_replicas, bool nonblock)
 700 {
 701         struct folio *folio;
 702         struct bch_folio *s;
 703         unsigned i, sectors;
 704         bool ret = true;
 705
 706         folio = __filemap_get_folio(mapping, *offset >> PAGE_SHIFT,
 707                                     FGP_LOCK|(nonblock ? FGP_NOWAIT : 0), 0);
 708         if (folio == ERR_PTR(-EAGAIN))
 709                 return -EAGAIN;
 710         if (IS_ERR_OR_NULL(folio))
 711                 return true;
 712
 713         s = bch2_folio(folio);
 714         if (!s)
 715                 goto unlock;
 716
 717         sectors = folio_sectors(folio);
 718         for (i = folio_pos_to_s(folio, *offset); i < sectors; i++)
 719                 if (s->s[i].state < SECTOR_dirty ||
 720                     s->s[i].nr_replicas + s->s[i].replicas_reserved < min_replicas) {
 721                         *offset = max(*offset,
 722                                       folio_pos(folio) + (i << SECTOR_SHIFT));
 723                         goto unlock;
 724                 }
 725
 726         *offset = folio_end_pos(folio);
 727         ret = false;
 728 unlock:
 729         folio_unlock(folio);
 730         folio_put(folio);
 731         return ret;
 732 }
 733
 734 loff_t bch2_seek_pagecache_hole(struct inode *vinode,
 735                                 loff_t start_offset,
 736                                 loff_t end_offset,
 737                                 unsigned min_replicas,
 738                                 bool nonblock)
 739 {
 740         struct address_space *mapping = vinode->i_mapping;
 741         loff_t offset = start_offset;
 742
 743         while (offset < end_offset &&
 744                !folio_hole_offset(mapping, &offset, min_replicas, nonblock))
 745                 ;
 746
 747         return min(offset, end_offset);
 748 }
 749
 750 int bch2_clamp_data_hole(struct inode *inode,
 751                          u64 *hole_start,
 752                          u64 *hole_end,
 753                          unsigned min_replicas,
 754                          bool nonblock)
 755 {
 756         loff_t ret;
 757
 758         ret = bch2_seek_pagecache_hole(inode,
 759                 *hole_start << 9, *hole_end << 9, min_replicas, nonblock) >> 9;
 760         if (ret < 0)
 761                 return ret;
 762
 763         *hole_start = ret;
 764
 765         if (*hole_start == *hole_end)
 766                 return 0;
 767
 768         ret = bch2_seek_pagecache_data(inode,
 769                 *hole_start << 9, *hole_end << 9, min_replicas, nonblock) >> 9;
 770         if (ret < 0)
 771                 return ret;
 772
 773         *hole_end = ret;
 774         return 0;
 775 }
 776
 777 #endif /* NO_BCACHEFS_FS */